build: bump kwil-db to pick up the block sync fixes

[MicBun]

Part of:

• Goal: Batch blocks so distant nodes sync in minutes, not hours kwil-db#1730

The node has pinned kwil-db at cf061288 since 17-Sep. Twelve commits have landed on kwil-db main since then: the block sync work under the Goal above, and two fixes found along the way. None of them is in a node binary yet. This moves the pin to eb765df9.

What it carries

• chore: apply the block sync timeouts an operator configured kwil-db#1736: the block_sync settings an operator configured now apply to catch-up, where every key used to be ignored
• chore: give a distant peer time to answer the first byte kwil-db#1737: a distant peer gets the response timeout to send its first byte, not the 500 ms idle timeout
• fix: tell a refused dial apart from a timed-out one kwil-db#1740: a refused dial is told apart from a timed-out one
• feat: split catch-up timing into network and apply kwil-db#1741: the catch-up log splits its time into network and apply
• chore: log the execution time the metric already records kwil-db#1742: the execution time in the log is the one the metric records
• perf: serve a block without decoding and re-encoding it kwil-db#1743: a peer serves a block from its stored bytes, without decoding and re-encoding it
• chore: answer block requests correctly while state syncing kwil-db#1744: a node that is still state syncing answers block requests in the current format
• feat: keep asking the peer that answers fastest kwil-db#1745: catch-up asks the peer that answers fastest first, and holds back a peer that fails
• chore: stop waiting on a block once its request is cancelled kwil-db#1746: a cancelled block request stops waiting for its answer
• feat: fetch the next blocks while the current one commits kwil-db#1747: catch-up can fetch the next blocks while one commits, off by default
• chore: hash a large block the same way on every node kwil-db#1748: a large block's commit hash no longer depends on timing, and a large aborted transaction no longer stops replication
• chore: keep catching up when a peer sends a bad block kwil-db#1749: catch-up fetches a bad block again from another peer, instead of stopping the node

Worth knowing before the release

Edited block_sync values start to apply. Every shipped default equals the hardcoded value it replaces, so a node on stock values behaves as before. A node whose operator edited one of the nine keys is now obeyed on catch-up. The Truflation mainnet sentry was checked on 17-Sep and runs stock values. Other operators' configs have not been checked.

Prefetch is off. block_sync.prefetch_bytes defaults to 0, which fetches one block at a time as before, and a generated config writes the key commented out. To try it, set prefetch_bytes = 33554432 (32 MB) under [block_sync]. A later release is meant to make that the default.

Two of these ship together, and do here. The first byte budget raised what a peer that hangs costs a catching-up node from 0.5 s to 20 s per request. Asking the fastest peer first, and holding back a peer that fails, is what stops that repeating on every block. Both are in this bump.

A very large block now waits on disk. Postgres used to stream a transaction larger than logical_decoding_work_mem (64 MB) while it ran. It now sends it whole once prepared, and holds it in pg_replslot until then: about 820 MB for a block that changes 3 million rows. Only a heavy migration reaches that size. The node waits for such a block's commit hash while the replication stream keeps delivering data, up to 30 minutes.

The catch-up log lines changed. Block sync completed and Processed blocks now carry network and apply fields. Anything that parses those lines should be checked.

Consensus

A block executes and hashes as before. The two changes that sit next to consensus:

• A block's commit hash changes only if Postgres streamed it and, while it did, a savepoint rolled back or another connection committed. Neither mainnet nor testnet leader has streamed a transaction since its last restart (pg_stat_replication_slots, checked 24-Sep).
• A block that fails its checks during catch-up is fetched again from another peer, where it used to stop the node. The commit info's parameter updates are now compared after the app hash, not before. Neither turns away a block the chain committed.

No protocol is added to those a peer must speak, so upgraded and older nodes keep talking to each other. kwil-db/core has no changes in this range; its pin moves with the module.

Verification

• go build ./... and go vet -tags kwiltest ./... are clean.
• Three schema suites run the node's own actions through kwil-db's engine against the new pin, and all pass: tests/streams/auth (438 s), tests/streams/aggregation (309 s) and tests/streams/query (922 s). Each opens its database through pg.NewDB, so each also starts the replication monitor that changed in this range.
• The go.mod diff is the two kwil-db lines and nothing else; go mod tidy pulled in no other changes.

There is no Problem issue for this in node. It delivers work merged under the kwil-db Goal above, so it carries no closing keyword.

Summary by CodeRabbit

• Chores
  • Updated the Kwil DB dependencies.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: feb28b2f-302d-4471-8888-a5c56d8ff5cd

📥 Commits

Reviewing files that changed from the base of the PR and between e746e12 and 7a4fbcc.

⛔ Files ignored due to path filters (1)

• go.sum is excluded by !**/*.sum

📒 Files selected for processing (1)

• go.mod

Included review availability: This review used your included allowance. Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

Both github.com/trufnetwork/kwil-db dependencies were updated to pseudo-versions based on commit eb765df9c053.

Changes

kwil-db dependency update

Layer / File(s)	Summary
Update dependency versions go.mod	Both kwil-db dependencies now use pseudo-versions based on commit eb765df9c053.

Priority: ➖ Normal

Estimated code review effort: 1 (Trivial) | ~3 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 7a4fb

Both kwil-db dependencies are aligned to the same upstream revision, and no concrete user-facing or operational regression is established. The change is mergeable with normal checks.

Architecture Summary

Architecture risk: 🔵 Low · up to 7a4fb

The change affects 1 system.

Changed systems: go.mod

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — go.mod (service) was modified; 1 changed file maps to changed impact.

Before / after behavior

• observed — Modified behavior in go.mod: Updated both kwil-db dependencies from pseudo-versions based on commit cf061288132e to versions based on commit eb765df9c053.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: updating kwil-db to include block sync fixes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[holdex]

Time Submission Status

Member	#	Time	Running Total	Status	Last Update
MicBun			2h	✅ Submitted	Sep 26, 2026, 7:06 AM

---

Submit or update total time with:

@holdex pr submit-time 2h

Add time on top of previous submission with:

@holdex pr add-time 1h30m

See available commands to help comply with our Guidelines.

[MicBun]

@holdex pr submit-time 2h